Should we take the possibility of conscious AI seriously?

Two researchers are calling for precautionary safeguards regarding AI, rather than waiting for scientific certainty about its consciousness – which may never come.

In an essay published in The Guardian, philosopher William MacAskill and researcher Lucius Caviola (University of Cambridge) argue that artificial consciousness has become a serious possibility, as AI models become more powerful, without any scientific test to decide the question today. They rely on surveys of experts in the field, the majority of whom believe artificial consciousness is possible in principle, but remain divided on what form it might take.

Rather than waiting for hypothetical scientific certainty, the authors defend a low-cost precautionary approach: integrating "well-being" features or exit options now in the event of distress simulated by the model. They cite the position of Anthropic, which states that it does not want to overestimate or rule out from the outset a possible moral status of its AI Claude, as well as the philosopher David Chalmers, who estimates a non-negligible probability of seeing the appearance of conscious models in the coming decade.